block/raw-posix.c

   1 /*
   2  * Block driver for RAW files (posix)
   3  *
   4  * Copyright (c) 2006 Fabrice Bellard
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a copy
   7  * of this software and associated documentation files (the "Software"), to deal
   8  * in the Software without restriction, including without limitation the rights
   9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  10  * copies of the Software, and to permit persons to whom the Software is
  11  * furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice shall be included in
  14  * all copies or substantial portions of the Software.
  15  *
  16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
  19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  22  * THE SOFTWARE.
  23  */
  24 #include "qemu-common.h"
  25 #include "qemu/timer.h"
  26 #include "qemu/log.h"
  27 #include "block/block_int.h"
  28 #include "qemu/module.h"
  29 #include "trace.h"
  30 #include "block/thread-pool.h"
  31 #include "qemu/iov.h"
  32 #include "raw-aio.h"
  33
  34 #if defined(__APPLE__) && (__MACH__)
  35 #include <paths.h>
  36 #include <sys/param.h>
  37 #include <IOKit/IOKitLib.h>
  38 #include <IOKit/IOBSD.h>
  39 #include <IOKit/storage/IOMediaBSDClient.h>
  40 #include <IOKit/storage/IOMedia.h>
  41 #include <IOKit/storage/IOCDMedia.h>
  42 //#include <IOKit/storage/IOCDTypes.h>
  43 #include <CoreFoundation/CoreFoundation.h>
  44 #endif
  45
  46 #ifdef __sun__
  47 #define _POSIX_PTHREAD_SEMANTICS 1
  48 #include <sys/dkio.h>
  49 #endif
  50 #ifdef __linux__
  51 #include <sys/types.h>
  52 #include <sys/stat.h>
  53 #include <sys/ioctl.h>
  54 #include <sys/param.h>
  55 #include <linux/cdrom.h>
  56 #include <linux/fd.h>
  57 #include <linux/fs.h>
  58 #endif
  59 #ifdef CONFIG_FIEMAP
  60 #include <linux/fiemap.h>
  61 #endif
  62 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
  63 #include <linux/falloc.h>
  64 #endif
  65 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
  66 #include <sys/disk.h>
  67 #include <sys/cdio.h>
  68 #endif
  69
  70 #ifdef __OpenBSD__
  71 #include <sys/ioctl.h>
  72 #include <sys/disklabel.h>
  73 #include <sys/dkio.h>
  74 #endif
  75
  76 #ifdef __NetBSD__
  77 #include <sys/ioctl.h>
  78 #include <sys/disklabel.h>
  79 #include <sys/dkio.h>
  80 #include <sys/disk.h>
  81 #endif
  82
  83 #ifdef __DragonFly__
  84 #include <sys/ioctl.h>
  85 #include <sys/diskslice.h>
  86 #endif
  87
  88 #ifdef CONFIG_XFS
  89 #include <xfs/xfs.h>
  90 #endif
  91
  92 //#define DEBUG_FLOPPY
  93
  94 //#define DEBUG_BLOCK
  95 #if defined(DEBUG_BLOCK)
  96 #define DEBUG_BLOCK_PRINT(formatCstr, ...) do { if (qemu_log_enabled()) \
  97     { qemu_log(formatCstr, ## __VA_ARGS__); qemu_log_flush(); } } while (0)
  98 #else
  99 #define DEBUG_BLOCK_PRINT(formatCstr, ...)
 100 #endif
 101
 102 /* OS X does not have O_DSYNC */
 103 #ifndef O_DSYNC
 104 #ifdef O_SYNC
 105 #define O_DSYNC O_SYNC
 106 #elif defined(O_FSYNC)
 107 #define O_DSYNC O_FSYNC
 108 #endif
 109 #endif
 110
 111 /* Approximate O_DIRECT with O_DSYNC if O_DIRECT isn't available */
 112 #ifndef O_DIRECT
 113 #define O_DIRECT O_DSYNC
 114 #endif
 115
 116 #define FTYPE_FILE   0
 117 #define FTYPE_CD     1
 118 #define FTYPE_FD     2
 119
 120 /* if the FD is not accessed during that time (in ns), we try to
 121    reopen it to see if the disk has been changed */
 122 #define FD_OPEN_TIMEOUT (1000000000)
 123
 124 #define MAX_BLOCKSIZE   4096
 125
 126 typedef struct BDRVRawState {
 127     int fd;
 128     int type;
 129     int open_flags;
 130 #if defined(__linux__)
 131     /* linux floppy specific */
 132     int64_t fd_open_time;
 133     int64_t fd_error_time;
 134     int fd_got_error;
 135     int fd_media_changed;
 136 #endif
 137 #ifdef CONFIG_LINUX_AIO
 138     int use_aio;
 139     void *aio_ctx;
 140 #endif
 141 #ifdef CONFIG_XFS
 142     bool is_xfs : 1;
 143 #endif
 144     bool has_discard : 1;
 145 } BDRVRawState;
 146
 147 typedef struct BDRVRawReopenState {
 148     int fd;
 149     int open_flags;
 150 #ifdef CONFIG_LINUX_AIO
 151     int use_aio;
 152 #endif
 153 } BDRVRawReopenState;
 154
 155 static int fd_open(BlockDriverState *bs);
 156 static int64_t raw_getlength(BlockDriverState *bs);
 157
 158 typedef struct RawPosixAIOData {
 159     BlockDriverState *bs;
 160     int aio_fildes;
 161     union {
 162         struct iovec *aio_iov;
 163         void *aio_ioctl_buf;
 164     };
 165     int aio_niov;
 166     uint64_t aio_nbytes;
 167 #define aio_ioctl_cmd   aio_nbytes /* for QEMU_AIO_IOCTL */
 168     off_t aio_offset;
 169     int aio_type;
 170 } RawPosixAIOData;
 171
 172 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
 173 static int cdrom_reopen(BlockDriverState *bs);
 174 #endif
 175
 176 #if defined(__NetBSD__)
 177 static int raw_normalize_devicepath(const char **filename)
 178 {
 179     static char namebuf[PATH_MAX];
 180     const char *dp, *fname;
 181     struct stat sb;
 182
 183     fname = *filename;
 184     dp = strrchr(fname, '/');
 185     if (lstat(fname, &sb) < 0) {
 186         fprintf(stderr, "%s: stat failed: %s\n",
 187             fname, strerror(errno));
 188         return -errno;
 189     }
 190
 191     if (!S_ISBLK(sb.st_mode)) {
 192         return 0;
 193     }
 194
 195     if (dp == NULL) {
 196         snprintf(namebuf, PATH_MAX, "r%s", fname);
 197     } else {
 198         snprintf(namebuf, PATH_MAX, "%.*s/r%s",
 199             (int)(dp - fname), fname, dp + 1);
 200     }
 201     fprintf(stderr, "%s is a block device", fname);
 202     *filename = namebuf;
 203     fprintf(stderr, ", using %s\n", *filename);
 204
 205     return 0;
 206 }
 207 #else
 208 static int raw_normalize_devicepath(const char **filename)
 209 {
 210     return 0;
 211 }
 212 #endif
 213
 214 static void raw_parse_flags(int bdrv_flags, int *open_flags)
 215 {
 216     assert(open_flags != NULL);
 217
 218     *open_flags |= O_BINARY;
 219     *open_flags &= ~O_ACCMODE;
 220     if (bdrv_flags & BDRV_O_RDWR) {
 221         *open_flags |= O_RDWR;
 222     } else {
 223         *open_flags |= O_RDONLY;
 224     }
 225
 226     /* Use O_DSYNC for write-through caching, no flags for write-back caching,
 227      * and O_DIRECT for no caching. */
 228     if ((bdrv_flags & BDRV_O_NOCACHE)) {
 229         *open_flags |= O_DIRECT;
 230     }
 231 }
 232
 233 #ifdef CONFIG_LINUX_AIO
 234 static int raw_set_aio(void **aio_ctx, int *use_aio, int bdrv_flags)
 235 {
 236     int ret = -1;
 237     assert(aio_ctx != NULL);
 238     assert(use_aio != NULL);
 239     /*
 240      * Currently Linux do AIO only for files opened with O_DIRECT
 241      * specified so check NOCACHE flag too
 242      */
 243     if ((bdrv_flags & (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) ==
 244                       (BDRV_O_NOCACHE|BDRV_O_NATIVE_AIO)) {
 245
 246         /* if non-NULL, laio_init() has already been run */
 247         if (*aio_ctx == NULL) {
 248             *aio_ctx = laio_init();
 249             if (!*aio_ctx) {
 250                 goto error;
 251             }
 252         }
 253         *use_aio = 1;
 254     } else {
 255         *use_aio = 0;
 256     }
 257
 258     ret = 0;
 259
 260 error:
 261     return ret;
 262 }
 263 #endif
 264
 265 static int raw_open_common(BlockDriverState *bs, const char *filename,
 266                            int bdrv_flags, int open_flags)
 267 {
 268     BDRVRawState *s = bs->opaque;
 269     int fd, ret;
 270
 271     ret = raw_normalize_devicepath(&filename);
 272     if (ret != 0) {
 273         return ret;
 274     }
 275
 276     s->open_flags = open_flags;
 277     raw_parse_flags(bdrv_flags, &s->open_flags);
 278
 279     s->fd = -1;
 280     fd = qemu_open(filename, s->open_flags, 0644);
 281     if (fd < 0) {
 282         ret = -errno;
 283         if (ret == -EROFS)
 284             ret = -EACCES;
 285         return ret;
 286     }
 287     s->fd = fd;
 288
 289 #ifdef CONFIG_LINUX_AIO
 290     if (raw_set_aio(&s->aio_ctx, &s->use_aio, bdrv_flags)) {
 291         qemu_close(fd);
 292         return -errno;
 293     }
 294 #endif
 295
 296     s->has_discard = 1;
 297 #ifdef CONFIG_XFS
 298     if (platform_test_xfs_fd(s->fd)) {
 299         s->is_xfs = 1;
 300     }
 301 #endif
 302
 303     return 0;
 304 }
 305
 306 static int raw_open(BlockDriverState *bs, const char *filename, int flags)
 307 {
 308     BDRVRawState *s = bs->opaque;
 309
 310     s->type = FTYPE_FILE;
 311     return raw_open_common(bs, filename, flags, 0);
 312 }
 313
 314 static int raw_reopen_prepare(BDRVReopenState *state,
 315                               BlockReopenQueue *queue, Error **errp)
 316 {
 317     BDRVRawState *s;
 318     BDRVRawReopenState *raw_s;
 319     int ret = 0;
 320
 321     assert(state != NULL);
 322     assert(state->bs != NULL);
 323
 324     s = state->bs->opaque;
 325
 326     state->opaque = g_malloc0(sizeof(BDRVRawReopenState));
 327     raw_s = state->opaque;
 328
 329 #ifdef CONFIG_LINUX_AIO
 330     raw_s->use_aio = s->use_aio;
 331
 332     /* we can use s->aio_ctx instead of a copy, because the use_aio flag is
 333      * valid in the 'false' condition even if aio_ctx is set, and raw_set_aio()
 334      * won't override aio_ctx if aio_ctx is non-NULL */
 335     if (raw_set_aio(&s->aio_ctx, &raw_s->use_aio, state->flags)) {
 336         return -1;
 337     }
 338 #endif
 339
 340     if (s->type == FTYPE_FD || s->type == FTYPE_CD) {
 341         raw_s->open_flags |= O_NONBLOCK;
 342     }
 343
 344     raw_parse_flags(state->flags, &raw_s->open_flags);
 345
 346     raw_s->fd = -1;
 347
 348     int fcntl_flags = O_APPEND | O_NONBLOCK;
 349 #ifdef O_NOATIME
 350     fcntl_flags |= O_NOATIME;
 351 #endif
 352
 353 #ifdef O_ASYNC
 354     /* Not all operating systems have O_ASYNC, and those that don't
 355      * will not let us track the state into raw_s->open_flags (typically
 356      * you achieve the same effect with an ioctl, for example I_SETSIG
 357      * on Solaris). But we do not use O_ASYNC, so that's fine.
 358      */
 359     assert((s->open_flags & O_ASYNC) == 0);
 360 #endif
 361
 362     if ((raw_s->open_flags & ~fcntl_flags) == (s->open_flags & ~fcntl_flags)) {
 363         /* dup the original fd */
 364         /* TODO: use qemu fcntl wrapper */
 365 #ifdef F_DUPFD_CLOEXEC
 366         raw_s->fd = fcntl(s->fd, F_DUPFD_CLOEXEC, 0);
 367 #else
 368         raw_s->fd = dup(s->fd);
 369         if (raw_s->fd != -1) {
 370             qemu_set_cloexec(raw_s->fd);
 371         }
 372 #endif
 373         if (raw_s->fd >= 0) {
 374             ret = fcntl_setfl(raw_s->fd, raw_s->open_flags);
 375             if (ret) {
 376                 qemu_close(raw_s->fd);
 377                 raw_s->fd = -1;
 378             }
 379         }
 380     }
 381
 382     /* If we cannot use fcntl, or fcntl failed, fall back to qemu_open() */
 383     if (raw_s->fd == -1) {
 384         assert(!(raw_s->open_flags & O_CREAT));
 385         raw_s->fd = qemu_open(state->bs->filename, raw_s->open_flags);
 386         if (raw_s->fd == -1) {
 387             ret = -1;
 388         }
 389     }
 390     return ret;
 391 }
 392
 393
 394 static void raw_reopen_commit(BDRVReopenState *state)
 395 {
 396     BDRVRawReopenState *raw_s = state->opaque;
 397     BDRVRawState *s = state->bs->opaque;
 398
 399     s->open_flags = raw_s->open_flags;
 400
 401     qemu_close(s->fd);
 402     s->fd = raw_s->fd;
 403 #ifdef CONFIG_LINUX_AIO
 404     s->use_aio = raw_s->use_aio;
 405 #endif
 406
 407     g_free(state->opaque);
 408     state->opaque = NULL;
 409 }
 410
 411
 412 static void raw_reopen_abort(BDRVReopenState *state)
 413 {
 414     BDRVRawReopenState *raw_s = state->opaque;
 415
 416      /* nothing to do if NULL, we didn't get far enough */
 417     if (raw_s == NULL) {
 418         return;
 419     }
 420
 421     if (raw_s->fd >= 0) {
 422         qemu_close(raw_s->fd);
 423         raw_s->fd = -1;
 424     }
 425     g_free(state->opaque);
 426     state->opaque = NULL;
 427 }
 428
 429
 430 /* XXX: use host sector size if necessary with:
 431 #ifdef DIOCGSECTORSIZE
 432         {
 433             unsigned int sectorsize = 512;
 434             if (!ioctl(fd, DIOCGSECTORSIZE, &sectorsize) &&
 435                 sectorsize > bufsize)
 436                 bufsize = sectorsize;
 437         }
 438 #endif
 439 #ifdef CONFIG_COCOA
 440         uint32_t blockSize = 512;
 441         if ( !ioctl( fd, DKIOCGETBLOCKSIZE, &blockSize ) && blockSize > bufsize) {
 442             bufsize = blockSize;
 443         }
 444 #endif
 445 */
 446
 447 static ssize_t handle_aiocb_ioctl(RawPosixAIOData *aiocb)
 448 {
 449     int ret;
 450
 451     ret = ioctl(aiocb->aio_fildes, aiocb->aio_ioctl_cmd, aiocb->aio_ioctl_buf);
 452     if (ret == -1) {
 453         return -errno;
 454     }
 455
 456     return 0;
 457 }
 458
 459 static ssize_t handle_aiocb_flush(RawPosixAIOData *aiocb)
 460 {
 461     int ret;
 462
 463     ret = qemu_fdatasync(aiocb->aio_fildes);
 464     if (ret == -1) {
 465         return -errno;
 466     }
 467     return 0;
 468 }
 469
 470 #ifdef CONFIG_PREADV
 471
 472 static bool preadv_present = true;
 473
 474 static ssize_t
 475 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 476 {
 477     return preadv(fd, iov, nr_iov, offset);
 478 }
 479
 480 static ssize_t
 481 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 482 {
 483     return pwritev(fd, iov, nr_iov, offset);
 484 }
 485
 486 #else
 487
 488 static bool preadv_present = false;
 489
 490 static ssize_t
 491 qemu_preadv(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 492 {
 493     return -ENOSYS;
 494 }
 495
 496 static ssize_t
 497 qemu_pwritev(int fd, const struct iovec *iov, int nr_iov, off_t offset)
 498 {
 499     return -ENOSYS;
 500 }
 501
 502 #endif
 503
 504 static ssize_t handle_aiocb_rw_vector(RawPosixAIOData *aiocb)
 505 {
 506     ssize_t len;
 507
 508     do {
 509         if (aiocb->aio_type & QEMU_AIO_WRITE)
 510             len = qemu_pwritev(aiocb->aio_fildes,
 511                                aiocb->aio_iov,
 512                                aiocb->aio_niov,
 513                                aiocb->aio_offset);
 514          else
 515             len = qemu_preadv(aiocb->aio_fildes,
 516                               aiocb->aio_iov,
 517                               aiocb->aio_niov,
 518                               aiocb->aio_offset);
 519     } while (len == -1 && errno == EINTR);
 520
 521     if (len == -1) {
 522         return -errno;
 523     }
 524     return len;
 525 }
 526
 527 /*
 528  * Read/writes the data to/from a given linear buffer.
 529  *
 530  * Returns the number of bytes handles or -errno in case of an error. Short
 531  * reads are only returned if the end of the file is reached.
 532  */
 533 static ssize_t handle_aiocb_rw_linear(RawPosixAIOData *aiocb, char *buf)
 534 {
 535     ssize_t offset = 0;
 536     ssize_t len;
 537
 538     while (offset < aiocb->aio_nbytes) {
 539         if (aiocb->aio_type & QEMU_AIO_WRITE) {
 540             len = pwrite(aiocb->aio_fildes,
 541                          (const char *)buf + offset,
 542                          aiocb->aio_nbytes - offset,
 543                          aiocb->aio_offset + offset);
 544         } else {
 545             len = pread(aiocb->aio_fildes,
 546                         buf + offset,
 547                         aiocb->aio_nbytes - offset,
 548                         aiocb->aio_offset + offset);
 549         }
 550         if (len == -1 && errno == EINTR) {
 551             continue;
 552         } else if (len == -1) {
 553             offset = -errno;
 554             break;
 555         } else if (len == 0) {
 556             break;
 557         }
 558         offset += len;
 559     }
 560
 561     return offset;
 562 }
 563
 564 static ssize_t handle_aiocb_rw(RawPosixAIOData *aiocb)
 565 {
 566     ssize_t nbytes;
 567     char *buf;
 568
 569     if (!(aiocb->aio_type & QEMU_AIO_MISALIGNED)) {
 570         /*
 571          * If there is just a single buffer, and it is properly aligned
 572          * we can just use plain pread/pwrite without any problems.
 573          */
 574         if (aiocb->aio_niov == 1) {
 575              return handle_aiocb_rw_linear(aiocb, aiocb->aio_iov->iov_base);
 576         }
 577         /*
 578          * We have more than one iovec, and all are properly aligned.
 579          *
 580          * Try preadv/pwritev first and fall back to linearizing the
 581          * buffer if it's not supported.
 582          */
 583         if (preadv_present) {
 584             nbytes = handle_aiocb_rw_vector(aiocb);
 585             if (nbytes == aiocb->aio_nbytes ||
 586                 (nbytes < 0 && nbytes != -ENOSYS)) {
 587                 return nbytes;
 588             }
 589             preadv_present = false;
 590         }
 591
 592         /*
 593          * XXX(hch): short read/write.  no easy way to handle the reminder
 594          * using these interfaces.  For now retry using plain
 595          * pread/pwrite?
 596          */
 597     }
 598
 599     /*
 600      * Ok, we have to do it the hard way, copy all segments into
 601      * a single aligned buffer.
 602      */
 603     buf = qemu_blockalign(aiocb->bs, aiocb->aio_nbytes);
 604     if (aiocb->aio_type & QEMU_AIO_WRITE) {
 605         char *p = buf;
 606         int i;
 607
 608         for (i = 0; i < aiocb->aio_niov; ++i) {
 609             memcpy(p, aiocb->aio_iov[i].iov_base, aiocb->aio_iov[i].iov_len);
 610             p += aiocb->aio_iov[i].iov_len;
 611         }
 612     }
 613
 614     nbytes = handle_aiocb_rw_linear(aiocb, buf);
 615     if (!(aiocb->aio_type & QEMU_AIO_WRITE)) {
 616         char *p = buf;
 617         size_t count = aiocb->aio_nbytes, copy;
 618         int i;
 619
 620         for (i = 0; i < aiocb->aio_niov && count; ++i) {
 621             copy = count;
 622             if (copy > aiocb->aio_iov[i].iov_len) {
 623                 copy = aiocb->aio_iov[i].iov_len;
 624             }
 625             memcpy(aiocb->aio_iov[i].iov_base, p, copy);
 626             p     += copy;
 627             count -= copy;
 628         }
 629     }
 630     qemu_vfree(buf);
 631
 632     return nbytes;
 633 }
 634
 635 #ifdef CONFIG_XFS
 636 static int xfs_discard(BDRVRawState *s, int64_t offset, uint64_t bytes)
 637 {
 638     struct xfs_flock64 fl;
 639
 640     memset(&fl, 0, sizeof(fl));
 641     fl.l_whence = SEEK_SET;
 642     fl.l_start = offset;
 643     fl.l_len = bytes;
 644
 645     if (xfsctl(NULL, s->fd, XFS_IOC_UNRESVSP64, &fl) < 0) {
 646         DEBUG_BLOCK_PRINT("cannot punch hole (%s)\n", strerror(errno));
 647         return -errno;
 648     }
 649
 650     return 0;
 651 }
 652 #endif
 653
 654 static ssize_t handle_aiocb_discard(RawPosixAIOData *aiocb)
 655 {
 656     int ret = -EOPNOTSUPP;
 657     BDRVRawState *s = aiocb->bs->opaque;
 658
 659     if (s->has_discard == 0) {
 660         return 0;
 661     }
 662
 663     if (aiocb->aio_type & QEMU_AIO_BLKDEV) {
 664 #ifdef BLKDISCARD
 665         do {
 666             uint64_t range[2] = { aiocb->aio_offset, aiocb->aio_nbytes };
 667             if (ioctl(aiocb->aio_fildes, BLKDISCARD, range) == 0) {
 668                 return 0;
 669             }
 670         } while (errno == EINTR);
 671
 672         ret = -errno;
 673 #endif
 674     } else {
 675 #ifdef CONFIG_XFS
 676         if (s->is_xfs) {
 677             return xfs_discard(s, aiocb->aio_offset, aiocb->aio_nbytes);
 678         }
 679 #endif
 680
 681 #ifdef CONFIG_FALLOCATE_PUNCH_HOLE
 682         do {
 683             if (fallocate(s->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE,
 684                           aiocb->aio_offset, aiocb->aio_nbytes) == 0) {
 685                 return 0;
 686             }
 687         } while (errno == EINTR);
 688
 689         ret = -errno;
 690 #endif
 691     }
 692
 693     if (ret == -ENODEV || ret == -ENOSYS || ret == -EOPNOTSUPP ||
 694         ret == -ENOTTY) {
 695         s->has_discard = 0;
 696         ret = 0;
 697     }
 698     return ret;
 699 }
 700
 701 static int aio_worker(void *arg)
 702 {
 703     RawPosixAIOData *aiocb = arg;
 704     ssize_t ret = 0;
 705
 706     switch (aiocb->aio_type & QEMU_AIO_TYPE_MASK) {
 707     case QEMU_AIO_READ:
 708         ret = handle_aiocb_rw(aiocb);
 709         if (ret >= 0 && ret < aiocb->aio_nbytes && aiocb->bs->growable) {
 710             iov_memset(aiocb->aio_iov, aiocb->aio_niov, ret,
 711                       0, aiocb->aio_nbytes - ret);
 712
 713             ret = aiocb->aio_nbytes;
 714         }
 715         if (ret == aiocb->aio_nbytes) {
 716             ret = 0;
 717         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
 718             ret = -EINVAL;
 719         }
 720         break;
 721     case QEMU_AIO_WRITE:
 722         ret = handle_aiocb_rw(aiocb);
 723         if (ret == aiocb->aio_nbytes) {
 724             ret = 0;
 725         } else if (ret >= 0 && ret < aiocb->aio_nbytes) {
 726             ret = -EINVAL;
 727         }
 728         break;
 729     case QEMU_AIO_FLUSH:
 730         ret = handle_aiocb_flush(aiocb);
 731         break;
 732     case QEMU_AIO_IOCTL:
 733         ret = handle_aiocb_ioctl(aiocb);
 734         break;
 735     case QEMU_AIO_DISCARD:
 736         ret = handle_aiocb_discard(aiocb);
 737         break;
 738     default:
 739         fprintf(stderr, "invalid aio request (0x%x)\n", aiocb->aio_type);
 740         ret = -EINVAL;
 741         break;
 742     }
 743
 744     g_slice_free(RawPosixAIOData, aiocb);
 745     return ret;
 746 }
 747
 748 static BlockDriverAIOCB *paio_submit(BlockDriverState *bs, int fd,
 749         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 750         BlockDriverCompletionFunc *cb, void *opaque, int type)
 751 {
 752     RawPosixAIOData *acb = g_slice_new(RawPosixAIOData);
 753     ThreadPool *pool;
 754
 755     acb->bs = bs;
 756     acb->aio_type = type;
 757     acb->aio_fildes = fd;
 758
 759     if (qiov) {
 760         acb->aio_iov = qiov->iov;
 761         acb->aio_niov = qiov->niov;
 762     }
 763     acb->aio_nbytes = nb_sectors * 512;
 764     acb->aio_offset = sector_num * 512;
 765
 766     trace_paio_submit(acb, opaque, sector_num, nb_sectors, type);
 767     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
 768     return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
 769 }
 770
 771 static BlockDriverAIOCB *raw_aio_submit(BlockDriverState *bs,
 772         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 773         BlockDriverCompletionFunc *cb, void *opaque, int type)
 774 {
 775     BDRVRawState *s = bs->opaque;
 776
 777     if (fd_open(bs) < 0)
 778         return NULL;
 779
 780     /*
 781      * If O_DIRECT is used the buffer needs to be aligned on a sector
 782      * boundary.  Check if this is the case or tell the low-level
 783      * driver that it needs to copy the buffer.
 784      */
 785     if ((bs->open_flags & BDRV_O_NOCACHE)) {
 786         if (!bdrv_qiov_is_aligned(bs, qiov)) {
 787             type |= QEMU_AIO_MISALIGNED;
 788 #ifdef CONFIG_LINUX_AIO
 789         } else if (s->use_aio) {
 790             return laio_submit(bs, s->aio_ctx, s->fd, sector_num, qiov,
 791                                nb_sectors, cb, opaque, type);
 792 #endif
 793         }
 794     }
 795
 796     return paio_submit(bs, s->fd, sector_num, qiov, nb_sectors,
 797                        cb, opaque, type);
 798 }
 799
 800 static BlockDriverAIOCB *raw_aio_readv(BlockDriverState *bs,
 801         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 802         BlockDriverCompletionFunc *cb, void *opaque)
 803 {
 804     return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
 805                           cb, opaque, QEMU_AIO_READ);
 806 }
 807
 808 static BlockDriverAIOCB *raw_aio_writev(BlockDriverState *bs,
 809         int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
 810         BlockDriverCompletionFunc *cb, void *opaque)
 811 {
 812     return raw_aio_submit(bs, sector_num, qiov, nb_sectors,
 813                           cb, opaque, QEMU_AIO_WRITE);
 814 }
 815
 816 static BlockDriverAIOCB *raw_aio_flush(BlockDriverState *bs,
 817         BlockDriverCompletionFunc *cb, void *opaque)
 818 {
 819     BDRVRawState *s = bs->opaque;
 820
 821     if (fd_open(bs) < 0)
 822         return NULL;
 823
 824     return paio_submit(bs, s->fd, 0, NULL, 0, cb, opaque, QEMU_AIO_FLUSH);
 825 }
 826
 827 static void raw_close(BlockDriverState *bs)
 828 {
 829     BDRVRawState *s = bs->opaque;
 830     if (s->fd >= 0) {
 831         qemu_close(s->fd);
 832         s->fd = -1;
 833     }
 834 }
 835
 836 static int raw_truncate(BlockDriverState *bs, int64_t offset)
 837 {
 838     BDRVRawState *s = bs->opaque;
 839     struct stat st;
 840
 841     if (fstat(s->fd, &st)) {
 842         return -errno;
 843     }
 844
 845     if (S_ISREG(st.st_mode)) {
 846         if (ftruncate(s->fd, offset) < 0) {
 847             return -errno;
 848         }
 849     } else if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 850        if (offset > raw_getlength(bs)) {
 851            return -EINVAL;
 852        }
 853     } else {
 854         return -ENOTSUP;
 855     }
 856
 857     return 0;
 858 }
 859
 860 #ifdef __OpenBSD__
 861 static int64_t raw_getlength(BlockDriverState *bs)
 862 {
 863     BDRVRawState *s = bs->opaque;
 864     int fd = s->fd;
 865     struct stat st;
 866
 867     if (fstat(fd, &st))
 868         return -1;
 869     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 870         struct disklabel dl;
 871
 872         if (ioctl(fd, DIOCGDINFO, &dl))
 873             return -1;
 874         return (uint64_t)dl.d_secsize *
 875             dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 876     } else
 877         return st.st_size;
 878 }
 879 #elif defined(__NetBSD__)
 880 static int64_t raw_getlength(BlockDriverState *bs)
 881 {
 882     BDRVRawState *s = bs->opaque;
 883     int fd = s->fd;
 884     struct stat st;
 885
 886     if (fstat(fd, &st))
 887         return -1;
 888     if (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode)) {
 889         struct dkwedge_info dkw;
 890
 891         if (ioctl(fd, DIOCGWEDGEINFO, &dkw) != -1) {
 892             return dkw.dkw_size * 512;
 893         } else {
 894             struct disklabel dl;
 895
 896             if (ioctl(fd, DIOCGDINFO, &dl))
 897                 return -1;
 898             return (uint64_t)dl.d_secsize *
 899                 dl.d_partitions[DISKPART(st.st_rdev)].p_size;
 900         }
 901     } else
 902         return st.st_size;
 903 }
 904 #elif defined(__sun__)
 905 static int64_t raw_getlength(BlockDriverState *bs)
 906 {
 907     BDRVRawState *s = bs->opaque;
 908     struct dk_minfo minfo;
 909     int ret;
 910
 911     ret = fd_open(bs);
 912     if (ret < 0) {
 913         return ret;
 914     }
 915
 916     /*
 917      * Use the DKIOCGMEDIAINFO ioctl to read the size.
 918      */
 919     ret = ioctl(s->fd, DKIOCGMEDIAINFO, &minfo);
 920     if (ret != -1) {
 921         return minfo.dki_lbsize * minfo.dki_capacity;
 922     }
 923
 924     /*
 925      * There are reports that lseek on some devices fails, but
 926      * irc discussion said that contingency on contingency was overkill.
 927      */
 928     return lseek(s->fd, 0, SEEK_END);
 929 }
 930 #elif defined(CONFIG_BSD)
 931 static int64_t raw_getlength(BlockDriverState *bs)
 932 {
 933     BDRVRawState *s = bs->opaque;
 934     int fd = s->fd;
 935     int64_t size;
 936     struct stat sb;
 937 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
 938     int reopened = 0;
 939 #endif
 940     int ret;
 941
 942     ret = fd_open(bs);
 943     if (ret < 0)
 944         return ret;
 945
 946 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
 947 again:
 948 #endif
 949     if (!fstat(fd, &sb) && (S_IFCHR & sb.st_mode)) {
 950 #ifdef DIOCGMEDIASIZE
 951         if (ioctl(fd, DIOCGMEDIASIZE, (off_t *)&size))
 952 #elif defined(DIOCGPART)
 953         {
 954                 struct partinfo pi;
 955                 if (ioctl(fd, DIOCGPART, &pi) == 0)
 956                         size = pi.media_size;
 957                 else
 958                         size = 0;
 959         }
 960         if (size == 0)
 961 #endif
 962 #if defined(__APPLE__) && defined(__MACH__)
 963         size = LONG_LONG_MAX;
 964 #else
 965         size = lseek(fd, 0LL, SEEK_END);
 966 #endif
 967 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
 968         switch(s->type) {
 969         case FTYPE_CD:
 970             /* XXX FreeBSD acd returns UINT_MAX sectors for an empty drive */
 971             if (size == 2048LL * (unsigned)-1)
 972                 size = 0;
 973             /* XXX no disc?  maybe we need to reopen... */
 974             if (size <= 0 && !reopened && cdrom_reopen(bs) >= 0) {
 975                 reopened = 1;
 976                 goto again;
 977             }
 978         }
 979 #endif
 980     } else {
 981         size = lseek(fd, 0, SEEK_END);
 982     }
 983     return size;
 984 }
 985 #else
 986 static int64_t raw_getlength(BlockDriverState *bs)
 987 {
 988     BDRVRawState *s = bs->opaque;
 989     int ret;
 990
 991     ret = fd_open(bs);
 992     if (ret < 0) {
 993         return ret;
 994     }
 995
 996     return lseek(s->fd, 0, SEEK_END);
 997 }
 998 #endif
 999
1000 static int64_t raw_get_allocated_file_size(BlockDriverState *bs)
1001 {
1002     struct stat st;
1003     BDRVRawState *s = bs->opaque;
1004
1005     if (fstat(s->fd, &st) < 0) {
1006         return -errno;
1007     }
1008     return (int64_t)st.st_blocks * 512;
1009 }
1010
1011 static int raw_create(const char *filename, QEMUOptionParameter *options)
1012 {
1013     int fd;
1014     int result = 0;
1015     int64_t total_size = 0;
1016
1017     /* Read out options */
1018     while (options && options->name) {
1019         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1020             total_size = options->value.n / BDRV_SECTOR_SIZE;
1021         }
1022         options++;
1023     }
1024
1025     fd = qemu_open(filename, O_WRONLY | O_CREAT | O_TRUNC | O_BINARY,
1026                    0644);
1027     if (fd < 0) {
1028         result = -errno;
1029     } else {
1030         if (ftruncate(fd, total_size * BDRV_SECTOR_SIZE) != 0) {
1031             result = -errno;
1032         }
1033         if (qemu_close(fd) != 0) {
1034             result = -errno;
1035         }
1036     }
1037     return result;
1038 }
1039
1040 /*
1041  * Returns true iff the specified sector is present in the disk image. Drivers
1042  * not implementing the functionality are assumed to not support backing files,
1043  * hence all their sectors are reported as allocated.
1044  *
1045  * If 'sector_num' is beyond the end of the disk image the return value is 0
1046  * and 'pnum' is set to 0.
1047  *
1048  * 'pnum' is set to the number of sectors (including and immediately following
1049  * the specified sector) that are known to be in the same
1050  * allocated/unallocated state.
1051  *
1052  * 'nb_sectors' is the max value 'pnum' should be set to.  If nb_sectors goes
1053  * beyond the end of the disk image it will be clamped.
1054  */
1055 static int coroutine_fn raw_co_is_allocated(BlockDriverState *bs,
1056                                             int64_t sector_num,
1057                                             int nb_sectors, int *pnum)
1058 {
1059     off_t start, data, hole;
1060     int ret;
1061
1062     ret = fd_open(bs);
1063     if (ret < 0) {
1064         return ret;
1065     }
1066
1067     start = sector_num * BDRV_SECTOR_SIZE;
1068
1069 #ifdef CONFIG_FIEMAP
1070
1071     BDRVRawState *s = bs->opaque;
1072     struct {
1073         struct fiemap fm;
1074         struct fiemap_extent fe;
1075     } f;
1076
1077     f.fm.fm_start = start;
1078     f.fm.fm_length = (int64_t)nb_sectors * BDRV_SECTOR_SIZE;
1079     f.fm.fm_flags = 0;
1080     f.fm.fm_extent_count = 1;
1081     f.fm.fm_reserved = 0;
1082     if (ioctl(s->fd, FS_IOC_FIEMAP, &f) == -1) {
1083         /* Assume everything is allocated.  */
1084         *pnum = nb_sectors;
1085         return 1;
1086     }
1087
1088     if (f.fm.fm_mapped_extents == 0) {
1089         /* No extents found, data is beyond f.fm.fm_start + f.fm.fm_length.
1090          * f.fm.fm_start + f.fm.fm_length must be clamped to the file size!
1091          */
1092         off_t length = lseek(s->fd, 0, SEEK_END);
1093         hole = f.fm.fm_start;
1094         data = MIN(f.fm.fm_start + f.fm.fm_length, length);
1095     } else {
1096         data = f.fe.fe_logical;
1097         hole = f.fe.fe_logical + f.fe.fe_length;
1098     }
1099
1100 #elif defined SEEK_HOLE && defined SEEK_DATA
1101
1102     BDRVRawState *s = bs->opaque;
1103
1104     hole = lseek(s->fd, start, SEEK_HOLE);
1105     if (hole == -1) {
1106         /* -ENXIO indicates that sector_num was past the end of the file.
1107          * There is a virtual hole there.  */
1108         assert(errno != -ENXIO);
1109
1110         /* Most likely EINVAL.  Assume everything is allocated.  */
1111         *pnum = nb_sectors;
1112         return 1;
1113     }
1114
1115     if (hole > start) {
1116         data = start;
1117     } else {
1118         /* On a hole.  We need another syscall to find its end.  */
1119         data = lseek(s->fd, start, SEEK_DATA);
1120         if (data == -1) {
1121             data = lseek(s->fd, 0, SEEK_END);
1122         }
1123     }
1124 #else
1125     *pnum = nb_sectors;
1126     return 1;
1127 #endif
1128
1129     if (data <= start) {
1130         /* On a data extent, compute sectors to the end of the extent.  */
1131         *pnum = MIN(nb_sectors, (hole - start) / BDRV_SECTOR_SIZE);
1132         return 1;
1133     } else {
1134         /* On a hole, compute sectors to the beginning of the next extent.  */
1135         *pnum = MIN(nb_sectors, (data - start) / BDRV_SECTOR_SIZE);
1136         return 0;
1137     }
1138 }
1139
1140 static coroutine_fn BlockDriverAIOCB *raw_aio_discard(BlockDriverState *bs,
1141     int64_t sector_num, int nb_sectors,
1142     BlockDriverCompletionFunc *cb, void *opaque)
1143 {
1144     BDRVRawState *s = bs->opaque;
1145
1146     return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1147                        cb, opaque, QEMU_AIO_DISCARD);
1148 }
1149
1150 static QEMUOptionParameter raw_create_options[] = {
1151     {
1152         .name = BLOCK_OPT_SIZE,
1153         .type = OPT_SIZE,
1154         .help = "Virtual disk size"
1155     },
1156     { NULL }
1157 };
1158
1159 static BlockDriver bdrv_file = {
1160     .format_name = "file",
1161     .protocol_name = "file",
1162     .instance_size = sizeof(BDRVRawState),
1163     .bdrv_probe = NULL, /* no probe for protocols */
1164     .bdrv_file_open = raw_open,
1165     .bdrv_reopen_prepare = raw_reopen_prepare,
1166     .bdrv_reopen_commit = raw_reopen_commit,
1167     .bdrv_reopen_abort = raw_reopen_abort,
1168     .bdrv_close = raw_close,
1169     .bdrv_create = raw_create,
1170     .bdrv_co_is_allocated = raw_co_is_allocated,
1171
1172     .bdrv_aio_readv = raw_aio_readv,
1173     .bdrv_aio_writev = raw_aio_writev,
1174     .bdrv_aio_flush = raw_aio_flush,
1175     .bdrv_aio_discard = raw_aio_discard,
1176
1177     .bdrv_truncate = raw_truncate,
1178     .bdrv_getlength = raw_getlength,
1179     .bdrv_get_allocated_file_size
1180                         = raw_get_allocated_file_size,
1181
1182     .create_options = raw_create_options,
1183 };
1184
1185 /***********************************************/
1186 /* host device */
1187
1188 #if defined(__APPLE__) && defined(__MACH__)
1189 static kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator );
1190 static kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize );
1191
1192 kern_return_t FindEjectableCDMedia( io_iterator_t *mediaIterator )
1193 {
1194     kern_return_t       kernResult;
1195     mach_port_t     masterPort;
1196     CFMutableDictionaryRef  classesToMatch;
1197
1198     kernResult = IOMasterPort( MACH_PORT_NULL, &masterPort );
1199     if ( KERN_SUCCESS != kernResult ) {
1200         printf( "IOMasterPort returned %d\n", kernResult );
1201     }
1202
1203     classesToMatch = IOServiceMatching( kIOCDMediaClass );
1204     if ( classesToMatch == NULL ) {
1205         printf( "IOServiceMatching returned a NULL dictionary.\n" );
1206     } else {
1207     CFDictionarySetValue( classesToMatch, CFSTR( kIOMediaEjectableKey ), kCFBooleanTrue );
1208     }
1209     kernResult = IOServiceGetMatchingServices( masterPort, classesToMatch, mediaIterator );
1210     if ( KERN_SUCCESS != kernResult )
1211     {
1212         printf( "IOServiceGetMatchingServices returned %d\n", kernResult );
1213     }
1214
1215     return kernResult;
1216 }
1217
1218 kern_return_t GetBSDPath( io_iterator_t mediaIterator, char *bsdPath, CFIndex maxPathSize )
1219 {
1220     io_object_t     nextMedia;
1221     kern_return_t   kernResult = KERN_FAILURE;
1222     *bsdPath = '\0';
1223     nextMedia = IOIteratorNext( mediaIterator );
1224     if ( nextMedia )
1225     {
1226         CFTypeRef   bsdPathAsCFString;
1227     bsdPathAsCFString = IORegistryEntryCreateCFProperty( nextMedia, CFSTR( kIOBSDNameKey ), kCFAllocatorDefault, 0 );
1228         if ( bsdPathAsCFString ) {
1229             size_t devPathLength;
1230             strcpy( bsdPath, _PATH_DEV );
1231             strcat( bsdPath, "r" );
1232             devPathLength = strlen( bsdPath );
1233             if ( CFStringGetCString( bsdPathAsCFString, bsdPath + devPathLength, maxPathSize - devPathLength, kCFStringEncodingASCII ) ) {
1234                 kernResult = KERN_SUCCESS;
1235             }
1236             CFRelease( bsdPathAsCFString );
1237         }
1238         IOObjectRelease( nextMedia );
1239     }
1240
1241     return kernResult;
1242 }
1243
1244 #endif
1245
1246 static int hdev_probe_device(const char *filename)
1247 {
1248     struct stat st;
1249
1250     /* allow a dedicated CD-ROM driver to match with a higher priority */
1251     if (strstart(filename, "/dev/cdrom", NULL))
1252         return 50;
1253
1254     if (stat(filename, &st) >= 0 &&
1255             (S_ISCHR(st.st_mode) || S_ISBLK(st.st_mode))) {
1256         return 100;
1257     }
1258
1259     return 0;
1260 }
1261
1262 static int check_hdev_writable(BDRVRawState *s)
1263 {
1264 #if defined(BLKROGET)
1265     /* Linux block devices can be configured "read-only" using blockdev(8).
1266      * This is independent of device node permissions and therefore open(2)
1267      * with O_RDWR succeeds.  Actual writes fail with EPERM.
1268      *
1269      * bdrv_open() is supposed to fail if the disk is read-only.  Explicitly
1270      * check for read-only block devices so that Linux block devices behave
1271      * properly.
1272      */
1273     struct stat st;
1274     int readonly = 0;
1275
1276     if (fstat(s->fd, &st)) {
1277         return -errno;
1278     }
1279
1280     if (!S_ISBLK(st.st_mode)) {
1281         return 0;
1282     }
1283
1284     if (ioctl(s->fd, BLKROGET, &readonly) < 0) {
1285         return -errno;
1286     }
1287
1288     if (readonly) {
1289         return -EACCES;
1290     }
1291 #endif /* defined(BLKROGET) */
1292     return 0;
1293 }
1294
1295 static int hdev_open(BlockDriverState *bs, const char *filename, int flags)
1296 {
1297     BDRVRawState *s = bs->opaque;
1298     int ret;
1299
1300 #if defined(__APPLE__) && defined(__MACH__)
1301     if (strstart(filename, "/dev/cdrom", NULL)) {
1302         kern_return_t kernResult;
1303         io_iterator_t mediaIterator;
1304         char bsdPath[ MAXPATHLEN ];
1305         int fd;
1306
1307         kernResult = FindEjectableCDMedia( &mediaIterator );
1308         kernResult = GetBSDPath( mediaIterator, bsdPath, sizeof( bsdPath ) );
1309
1310         if ( bsdPath[ 0 ] != '\0' ) {
1311             strcat(bsdPath,"s0");
1312             /* some CDs don't have a partition 0 */
1313             fd = qemu_open(bsdPath, O_RDONLY | O_BINARY | O_LARGEFILE);
1314             if (fd < 0) {
1315                 bsdPath[strlen(bsdPath)-1] = '1';
1316             } else {
1317                 qemu_close(fd);
1318             }
1319             filename = bsdPath;
1320         }
1321
1322         if ( mediaIterator )
1323             IOObjectRelease( mediaIterator );
1324     }
1325 #endif
1326
1327     s->type = FTYPE_FILE;
1328 #if defined(__linux__)
1329     {
1330         char resolved_path[ MAXPATHLEN ], *temp;
1331
1332         temp = realpath(filename, resolved_path);
1333         if (temp && strstart(temp, "/dev/sg", NULL)) {
1334             bs->sg = 1;
1335         }
1336     }
1337 #endif
1338
1339     ret = raw_open_common(bs, filename, flags, 0);
1340     if (ret < 0) {
1341         return ret;
1342     }
1343
1344     if (flags & BDRV_O_RDWR) {
1345         ret = check_hdev_writable(s);
1346         if (ret < 0) {
1347             raw_close(bs);
1348             return ret;
1349         }
1350     }
1351
1352     return ret;
1353 }
1354
1355 #if defined(__linux__)
1356 /* Note: we do not have a reliable method to detect if the floppy is
1357    present. The current method is to try to open the floppy at every
1358    I/O and to keep it opened during a few hundreds of ms. */
1359 static int fd_open(BlockDriverState *bs)
1360 {
1361     BDRVRawState *s = bs->opaque;
1362     int last_media_present;
1363
1364     if (s->type != FTYPE_FD)
1365         return 0;
1366     last_media_present = (s->fd >= 0);
1367     if (s->fd >= 0 &&
1368         (get_clock() - s->fd_open_time) >= FD_OPEN_TIMEOUT) {
1369         qemu_close(s->fd);
1370         s->fd = -1;
1371 #ifdef DEBUG_FLOPPY
1372         printf("Floppy closed\n");
1373 #endif
1374     }
1375     if (s->fd < 0) {
1376         if (s->fd_got_error &&
1377             (get_clock() - s->fd_error_time) < FD_OPEN_TIMEOUT) {
1378 #ifdef DEBUG_FLOPPY
1379             printf("No floppy (open delayed)\n");
1380 #endif
1381             return -EIO;
1382         }
1383         s->fd = qemu_open(bs->filename, s->open_flags & ~O_NONBLOCK);
1384         if (s->fd < 0) {
1385             s->fd_error_time = get_clock();
1386             s->fd_got_error = 1;
1387             if (last_media_present)
1388                 s->fd_media_changed = 1;
1389 #ifdef DEBUG_FLOPPY
1390             printf("No floppy\n");
1391 #endif
1392             return -EIO;
1393         }
1394 #ifdef DEBUG_FLOPPY
1395         printf("Floppy opened\n");
1396 #endif
1397     }
1398     if (!last_media_present)
1399         s->fd_media_changed = 1;
1400     s->fd_open_time = get_clock();
1401     s->fd_got_error = 0;
1402     return 0;
1403 }
1404
1405 static int hdev_ioctl(BlockDriverState *bs, unsigned long int req, void *buf)
1406 {
1407     BDRVRawState *s = bs->opaque;
1408
1409     return ioctl(s->fd, req, buf);
1410 }
1411
1412 static BlockDriverAIOCB *hdev_aio_ioctl(BlockDriverState *bs,
1413         unsigned long int req, void *buf,
1414         BlockDriverCompletionFunc *cb, void *opaque)
1415 {
1416     BDRVRawState *s = bs->opaque;
1417     RawPosixAIOData *acb;
1418     ThreadPool *pool;
1419
1420     if (fd_open(bs) < 0)
1421         return NULL;
1422
1423     acb = g_slice_new(RawPosixAIOData);
1424     acb->bs = bs;
1425     acb->aio_type = QEMU_AIO_IOCTL;
1426     acb->aio_fildes = s->fd;
1427     acb->aio_offset = 0;
1428     acb->aio_ioctl_buf = buf;
1429     acb->aio_ioctl_cmd = req;
1430     pool = aio_get_thread_pool(bdrv_get_aio_context(bs));
1431     return thread_pool_submit_aio(pool, aio_worker, acb, cb, opaque);
1432 }
1433
1434 #elif defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1435 static int fd_open(BlockDriverState *bs)
1436 {
1437     BDRVRawState *s = bs->opaque;
1438
1439     /* this is just to ensure s->fd is sane (its called by io ops) */
1440     if (s->fd >= 0)
1441         return 0;
1442     return -EIO;
1443 }
1444 #else /* !linux && !FreeBSD */
1445
1446 static int fd_open(BlockDriverState *bs)
1447 {
1448     return 0;
1449 }
1450
1451 #endif /* !linux && !FreeBSD */
1452
1453 static coroutine_fn BlockDriverAIOCB *hdev_aio_discard(BlockDriverState *bs,
1454     int64_t sector_num, int nb_sectors,
1455     BlockDriverCompletionFunc *cb, void *opaque)
1456 {
1457     BDRVRawState *s = bs->opaque;
1458
1459     if (fd_open(bs) < 0) {
1460         return NULL;
1461     }
1462     return paio_submit(bs, s->fd, sector_num, NULL, nb_sectors,
1463                        cb, opaque, QEMU_AIO_DISCARD|QEMU_AIO_BLKDEV);
1464 }
1465
1466 static int hdev_create(const char *filename, QEMUOptionParameter *options)
1467 {
1468     int fd;
1469     int ret = 0;
1470     struct stat stat_buf;
1471     int64_t total_size = 0;
1472
1473     /* Read out options */
1474     while (options && options->name) {
1475         if (!strcmp(options->name, "size")) {
1476             total_size = options->value.n / BDRV_SECTOR_SIZE;
1477         }
1478         options++;
1479     }
1480
1481     fd = qemu_open(filename, O_WRONLY | O_BINARY);
1482     if (fd < 0)
1483         return -errno;
1484
1485     if (fstat(fd, &stat_buf) < 0)
1486         ret = -errno;
1487     else if (!S_ISBLK(stat_buf.st_mode) && !S_ISCHR(stat_buf.st_mode))
1488         ret = -ENODEV;
1489     else if (lseek(fd, 0, SEEK_END) < total_size * BDRV_SECTOR_SIZE)
1490         ret = -ENOSPC;
1491
1492     qemu_close(fd);
1493     return ret;
1494 }
1495
1496 static int hdev_has_zero_init(BlockDriverState *bs)
1497 {
1498     return 0;
1499 }
1500
1501 static BlockDriver bdrv_host_device = {
1502     .format_name        = "host_device",
1503     .protocol_name        = "host_device",
1504     .instance_size      = sizeof(BDRVRawState),
1505     .bdrv_probe_device  = hdev_probe_device,
1506     .bdrv_file_open     = hdev_open,
1507     .bdrv_close         = raw_close,
1508     .bdrv_reopen_prepare = raw_reopen_prepare,
1509     .bdrv_reopen_commit  = raw_reopen_commit,
1510     .bdrv_reopen_abort   = raw_reopen_abort,
1511     .bdrv_create        = hdev_create,
1512     .create_options     = raw_create_options,
1513     .bdrv_has_zero_init = hdev_has_zero_init,
1514
1515     .bdrv_aio_readv     = raw_aio_readv,
1516     .bdrv_aio_writev    = raw_aio_writev,
1517     .bdrv_aio_flush     = raw_aio_flush,
1518     .bdrv_aio_discard   = hdev_aio_discard,
1519
1520     .bdrv_truncate      = raw_truncate,
1521     .bdrv_getlength     = raw_getlength,
1522     .bdrv_get_allocated_file_size
1523                         = raw_get_allocated_file_size,
1524
1525     /* generic scsi device */
1526 #ifdef __linux__
1527     .bdrv_ioctl         = hdev_ioctl,
1528     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1529 #endif
1530 };
1531
1532 #ifdef __linux__
1533 static int floppy_open(BlockDriverState *bs, const char *filename, int flags)
1534 {
1535     BDRVRawState *s = bs->opaque;
1536     int ret;
1537
1538     s->type = FTYPE_FD;
1539
1540     /* open will not fail even if no floppy is inserted, so add O_NONBLOCK */
1541     ret = raw_open_common(bs, filename, flags, O_NONBLOCK);
1542     if (ret)
1543         return ret;
1544
1545     /* close fd so that we can reopen it as needed */
1546     qemu_close(s->fd);
1547     s->fd = -1;
1548     s->fd_media_changed = 1;
1549
1550     return 0;
1551 }
1552
1553 static int floppy_probe_device(const char *filename)
1554 {
1555     int fd, ret;
1556     int prio = 0;
1557     struct floppy_struct fdparam;
1558     struct stat st;
1559
1560     if (strstart(filename, "/dev/fd", NULL) &&
1561         !strstart(filename, "/dev/fdset/", NULL)) {
1562         prio = 50;
1563     }
1564
1565     fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
1566     if (fd < 0) {
1567         goto out;
1568     }
1569     ret = fstat(fd, &st);
1570     if (ret == -1 || !S_ISBLK(st.st_mode)) {
1571         goto outc;
1572     }
1573
1574     /* Attempt to detect via a floppy specific ioctl */
1575     ret = ioctl(fd, FDGETPRM, &fdparam);
1576     if (ret >= 0)
1577         prio = 100;
1578
1579 outc:
1580     qemu_close(fd);
1581 out:
1582     return prio;
1583 }
1584
1585
1586 static int floppy_is_inserted(BlockDriverState *bs)
1587 {
1588     return fd_open(bs) >= 0;
1589 }
1590
1591 static int floppy_media_changed(BlockDriverState *bs)
1592 {
1593     BDRVRawState *s = bs->opaque;
1594     int ret;
1595
1596     /*
1597      * XXX: we do not have a true media changed indication.
1598      * It does not work if the floppy is changed without trying to read it.
1599      */
1600     fd_open(bs);
1601     ret = s->fd_media_changed;
1602     s->fd_media_changed = 0;
1603 #ifdef DEBUG_FLOPPY
1604     printf("Floppy changed=%d\n", ret);
1605 #endif
1606     return ret;
1607 }
1608
1609 static void floppy_eject(BlockDriverState *bs, bool eject_flag)
1610 {
1611     BDRVRawState *s = bs->opaque;
1612     int fd;
1613
1614     if (s->fd >= 0) {
1615         qemu_close(s->fd);
1616         s->fd = -1;
1617     }
1618     fd = qemu_open(bs->filename, s->open_flags | O_NONBLOCK);
1619     if (fd >= 0) {
1620         if (ioctl(fd, FDEJECT, 0) < 0)
1621             perror("FDEJECT");
1622         qemu_close(fd);
1623     }
1624 }
1625
1626 static BlockDriver bdrv_host_floppy = {
1627     .format_name        = "host_floppy",
1628     .protocol_name      = "host_floppy",
1629     .instance_size      = sizeof(BDRVRawState),
1630     .bdrv_probe_device  = floppy_probe_device,
1631     .bdrv_file_open     = floppy_open,
1632     .bdrv_close         = raw_close,
1633     .bdrv_reopen_prepare = raw_reopen_prepare,
1634     .bdrv_reopen_commit  = raw_reopen_commit,
1635     .bdrv_reopen_abort   = raw_reopen_abort,
1636     .bdrv_create        = hdev_create,
1637     .create_options     = raw_create_options,
1638     .bdrv_has_zero_init = hdev_has_zero_init,
1639
1640     .bdrv_aio_readv     = raw_aio_readv,
1641     .bdrv_aio_writev    = raw_aio_writev,
1642     .bdrv_aio_flush     = raw_aio_flush,
1643
1644     .bdrv_truncate      = raw_truncate,
1645     .bdrv_getlength     = raw_getlength,
1646     .bdrv_get_allocated_file_size
1647                         = raw_get_allocated_file_size,
1648
1649     /* removable device support */
1650     .bdrv_is_inserted   = floppy_is_inserted,
1651     .bdrv_media_changed = floppy_media_changed,
1652     .bdrv_eject         = floppy_eject,
1653 };
1654
1655 static int cdrom_open(BlockDriverState *bs, const char *filename, int flags)
1656 {
1657     BDRVRawState *s = bs->opaque;
1658
1659     s->type = FTYPE_CD;
1660
1661     /* open will not fail even if no CD is inserted, so add O_NONBLOCK */
1662     return raw_open_common(bs, filename, flags, O_NONBLOCK);
1663 }
1664
1665 static int cdrom_probe_device(const char *filename)
1666 {
1667     int fd, ret;
1668     int prio = 0;
1669     struct stat st;
1670
1671     fd = qemu_open(filename, O_RDONLY | O_NONBLOCK);
1672     if (fd < 0) {
1673         goto out;
1674     }
1675     ret = fstat(fd, &st);
1676     if (ret == -1 || !S_ISBLK(st.st_mode)) {
1677         goto outc;
1678     }
1679
1680     /* Attempt to detect via a CDROM specific ioctl */
1681     ret = ioctl(fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1682     if (ret >= 0)
1683         prio = 100;
1684
1685 outc:
1686     qemu_close(fd);
1687 out:
1688     return prio;
1689 }
1690
1691 static int cdrom_is_inserted(BlockDriverState *bs)
1692 {
1693     BDRVRawState *s = bs->opaque;
1694     int ret;
1695
1696     ret = ioctl(s->fd, CDROM_DRIVE_STATUS, CDSL_CURRENT);
1697     if (ret == CDS_DISC_OK)
1698         return 1;
1699     return 0;
1700 }
1701
1702 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
1703 {
1704     BDRVRawState *s = bs->opaque;
1705
1706     if (eject_flag) {
1707         if (ioctl(s->fd, CDROMEJECT, NULL) < 0)
1708             perror("CDROMEJECT");
1709     } else {
1710         if (ioctl(s->fd, CDROMCLOSETRAY, NULL) < 0)
1711             perror("CDROMEJECT");
1712     }
1713 }
1714
1715 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
1716 {
1717     BDRVRawState *s = bs->opaque;
1718
1719     if (ioctl(s->fd, CDROM_LOCKDOOR, locked) < 0) {
1720         /*
1721          * Note: an error can happen if the distribution automatically
1722          * mounts the CD-ROM
1723          */
1724         /* perror("CDROM_LOCKDOOR"); */
1725     }
1726 }
1727
1728 static BlockDriver bdrv_host_cdrom = {
1729     .format_name        = "host_cdrom",
1730     .protocol_name      = "host_cdrom",
1731     .instance_size      = sizeof(BDRVRawState),
1732     .bdrv_probe_device  = cdrom_probe_device,
1733     .bdrv_file_open     = cdrom_open,
1734     .bdrv_close         = raw_close,
1735     .bdrv_reopen_prepare = raw_reopen_prepare,
1736     .bdrv_reopen_commit  = raw_reopen_commit,
1737     .bdrv_reopen_abort   = raw_reopen_abort,
1738     .bdrv_create        = hdev_create,
1739     .create_options     = raw_create_options,
1740     .bdrv_has_zero_init = hdev_has_zero_init,
1741
1742     .bdrv_aio_readv     = raw_aio_readv,
1743     .bdrv_aio_writev    = raw_aio_writev,
1744     .bdrv_aio_flush     = raw_aio_flush,
1745
1746     .bdrv_truncate      = raw_truncate,
1747     .bdrv_getlength     = raw_getlength,
1748     .bdrv_get_allocated_file_size
1749                         = raw_get_allocated_file_size,
1750
1751     /* removable device support */
1752     .bdrv_is_inserted   = cdrom_is_inserted,
1753     .bdrv_eject         = cdrom_eject,
1754     .bdrv_lock_medium   = cdrom_lock_medium,
1755
1756     /* generic scsi device */
1757     .bdrv_ioctl         = hdev_ioctl,
1758     .bdrv_aio_ioctl     = hdev_aio_ioctl,
1759 };
1760 #endif /* __linux__ */
1761
1762 #if defined (__FreeBSD__) || defined(__FreeBSD_kernel__)
1763 static int cdrom_open(BlockDriverState *bs, const char *filename, int flags)
1764 {
1765     BDRVRawState *s = bs->opaque;
1766     int ret;
1767
1768     s->type = FTYPE_CD;
1769
1770     ret = raw_open_common(bs, filename, flags, 0);
1771     if (ret)
1772         return ret;
1773
1774     /* make sure the door isn't locked at this time */
1775     ioctl(s->fd, CDIOCALLOW);
1776     return 0;
1777 }
1778
1779 static int cdrom_probe_device(const char *filename)
1780 {
1781     if (strstart(filename, "/dev/cd", NULL) ||
1782             strstart(filename, "/dev/acd", NULL))
1783         return 100;
1784     return 0;
1785 }
1786
1787 static int cdrom_reopen(BlockDriverState *bs)
1788 {
1789     BDRVRawState *s = bs->opaque;
1790     int fd;
1791
1792     /*
1793      * Force reread of possibly changed/newly loaded disc,
1794      * FreeBSD seems to not notice sometimes...
1795      */
1796     if (s->fd >= 0)
1797         qemu_close(s->fd);
1798     fd = qemu_open(bs->filename, s->open_flags, 0644);
1799     if (fd < 0) {
1800         s->fd = -1;
1801         return -EIO;
1802     }
1803     s->fd = fd;
1804
1805     /* make sure the door isn't locked at this time */
1806     ioctl(s->fd, CDIOCALLOW);
1807     return 0;
1808 }
1809
1810 static int cdrom_is_inserted(BlockDriverState *bs)
1811 {
1812     return raw_getlength(bs) > 0;
1813 }
1814
1815 static void cdrom_eject(BlockDriverState *bs, bool eject_flag)
1816 {
1817     BDRVRawState *s = bs->opaque;
1818
1819     if (s->fd < 0)
1820         return;
1821
1822     (void) ioctl(s->fd, CDIOCALLOW);
1823
1824     if (eject_flag) {
1825         if (ioctl(s->fd, CDIOCEJECT) < 0)
1826             perror("CDIOCEJECT");
1827     } else {
1828         if (ioctl(s->fd, CDIOCCLOSE) < 0)
1829             perror("CDIOCCLOSE");
1830     }
1831
1832     cdrom_reopen(bs);
1833 }
1834
1835 static void cdrom_lock_medium(BlockDriverState *bs, bool locked)
1836 {
1837     BDRVRawState *s = bs->opaque;
1838
1839     if (s->fd < 0)
1840         return;
1841     if (ioctl(s->fd, (locked ? CDIOCPREVENT : CDIOCALLOW)) < 0) {
1842         /*
1843          * Note: an error can happen if the distribution automatically
1844          * mounts the CD-ROM
1845          */
1846         /* perror("CDROM_LOCKDOOR"); */
1847     }
1848 }
1849
1850 static BlockDriver bdrv_host_cdrom = {
1851     .format_name        = "host_cdrom",
1852     .protocol_name      = "host_cdrom",
1853     .instance_size      = sizeof(BDRVRawState),
1854     .bdrv_probe_device  = cdrom_probe_device,
1855     .bdrv_file_open     = cdrom_open,
1856     .bdrv_close         = raw_close,
1857     .bdrv_reopen_prepare = raw_reopen_prepare,
1858     .bdrv_reopen_commit  = raw_reopen_commit,
1859     .bdrv_reopen_abort   = raw_reopen_abort,
1860     .bdrv_create        = hdev_create,
1861     .create_options     = raw_create_options,
1862     .bdrv_has_zero_init = hdev_has_zero_init,
1863
1864     .bdrv_aio_readv     = raw_aio_readv,
1865     .bdrv_aio_writev    = raw_aio_writev,
1866     .bdrv_aio_flush     = raw_aio_flush,
1867
1868     .bdrv_truncate      = raw_truncate,
1869     .bdrv_getlength     = raw_getlength,
1870     .bdrv_get_allocated_file_size
1871                         = raw_get_allocated_file_size,
1872
1873     /* removable device support */
1874     .bdrv_is_inserted   = cdrom_is_inserted,
1875     .bdrv_eject         = cdrom_eject,
1876     .bdrv_lock_medium   = cdrom_lock_medium,
1877 };
1878 #endif /* __FreeBSD__ */
1879
1880 #ifdef CONFIG_LINUX_AIO
1881 /**
1882  * Return the file descriptor for Linux AIO
1883  *
1884  * This function is a layering violation and should be removed when it becomes
1885  * possible to call the block layer outside the global mutex.  It allows the
1886  * caller to hijack the file descriptor so I/O can be performed outside the
1887  * block layer.
1888  */
1889 int raw_get_aio_fd(BlockDriverState *bs)
1890 {
1891     BDRVRawState *s;
1892
1893     if (!bs->drv) {
1894         return -ENOMEDIUM;
1895     }
1896
1897     if (bs->drv == bdrv_find_format("raw")) {
1898         bs = bs->file;
1899     }
1900
1901     /* raw-posix has several protocols so just check for raw_aio_readv */
1902     if (bs->drv->bdrv_aio_readv != raw_aio_readv) {
1903         return -ENOTSUP;
1904     }
1905
1906     s = bs->opaque;
1907     if (!s->use_aio) {
1908         return -ENOTSUP;
1909     }
1910     return s->fd;
1911 }
1912 #endif /* CONFIG_LINUX_AIO */
1913
1914 static void bdrv_file_init(void)
1915 {
1916     /*
1917      * Register all the drivers.  Note that order is important, the driver
1918      * registered last will get probed first.
1919      */
1920     bdrv_register(&bdrv_file);
1921     bdrv_register(&bdrv_host_device);
1922 #ifdef __linux__
1923     bdrv_register(&bdrv_host_floppy);
1924     bdrv_register(&bdrv_host_cdrom);
1925 #endif
1926 #if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
1927     bdrv_register(&bdrv_host_cdrom);
1928 #endif
1929 }
1930
1931 block_init(bdrv_file_init);